Selective Measurement of Glutamine and Asparagine in Aqueous

May 5, 1996 - Selective Measurement of Glutamine and Asparagine in Aqueous Media by ... Asparagine measurements were possible over the concentration ...
1 downloads 0 Views 1MB Size
Chapter 12

Selective Measurement of Glutamine and Asparagine in Aqueous Media by Near-Infrared Spectroscopy 1

1,3

1,4

2

Xiangji Zhou , Hoeil Chung , Mark A. Arnold , Martin Rhiel , and David W. Murhammer 2

1

2

Department of Chemistry and Department of Chemical and Biochemical Engineering, University of Iowa, Iowa City, IA 52242

Asparagine and glutamine concentrations have been determined in binary aqueous solutions with near-infrared (NIR) absorption spectroscopy. Spectra were collected over a range from 5000 to 4000 cm with a 1.0 mm optical path length cell. Differences in absorbance features around 4570 and 4380 cm for asparagine and glutamine provide the analytical information required for the resolution of these similar amino acids. The best overall performance was obtained by partial least-square (PLS) regression coupled with digital Fourier filtering over the spectral range of 4800-4250cm for asparagine and 4650-4320 cm for glutamine. Asparagine measurements were possible over the concentration range from 1.0 to 11.7 m M with a standard error of prediction (SEP) of 0.18 m M and a mean percent error of 2.50%. Glutamine could be measured over the concentration range from 1.0 to 13.6 m M with a SEP of 0.10 m M and a mean percent error of 2.00%. These results represent a critical first step in developing a NIR spectroscopic method for monitoring asparagine and glutamine in mammalian and insect cell cultures. -1

-1

-1

-1

Monitoring and control of processes are becoming increasingly important in the agricultural, pharmaceutical, textile, food and other industries (1, 16). For animal cell cultures, it is necessary to properly control the feeding of nutrients, removal of products and accumulation of by-product inhibitors in order to increase efficiency and reduce the cost of production (2-6). Enzyme-based biosensors (1), and NIR spectroscopy (7,10) are currently being developed as continuous monitors for bioreactors. Enzyme-based biosensors are 3

Current address: Chemical Process Research Lab, Yukong Limited, Ulsan, Korea Corresponding author

4

0097-6156/95/0613-0116$12.00/0 © 1995 American Chemical Society

12.

117 Selective Measurement of Glutamine and Asparagine

Z H O U E T AL.

not ideal for monitoring bioreactors. Although biosensors are relatively sensitive and can be selective for critical bioreactor analytes, such as glucose and amino acids, they are invasive in nature and are severely limited in terms of operational lifetime under conditions of most bioreactors. In addition, enzyme-based biosensors are difficult to sterilize and demand periodic calibration. In contrast, remote NIR spectroscopy is noninvasive and can, in principle, operate for extended periods of time without recalibration. A NIR approach is also capable of extracting quantitative information for many components from a single spectrum (7). The primary concern for NIR spectroscopy is the ability to accurately differentiate similar compounds given the rather broad and overlapping nature of NIR absorbance bands. Biologically important species, including amino acids, have unique spectral features which can be exploited to provide measurement selectivity. In the region of 5000 - 4000 cm" , the major absorption contributions come from combinations of vibrational transitions for aliphatic C-H, alkene C-H, amine N - H (ionized or not) and O-H bonds (8). Such spectral differences have been used to selectively measure structurally different compounds, such as glucose and ammonium ions (9), and glucose and glutamine (10). We have performed a study designed to establish the utility of NIR spectroscopy for distinguishing chemically similar compounds. Our test system is a set of binary mixtures of glutamine and asparagine in an aqueous buffer solution. These amino acids differ by a single methylene group in the side chain. The additional methylene group present in glutamine affects the spectral absorbance feature around 4400 cm" which is predominately composed of combination bands of stretching and bending transitions of the C - H bonds (11). Individual absorbance spectra are presented in Figure 1 for asparagine and glutamine. Comparison of these spectra reveals small differences in the 4400 cm" region. Our investigation has established that these small spectral differences are sufficient to differentiate glutamine and asparagine at the millimolar concentration level. PLS regression coupled to digital Fourier filtering has been used for these measurements. 1

1

1

Experimental Apparatus, Reagents and Solutions. A Nicolet 740 Fourier transform infrared (FTIR) spectrometer (Nicolet Analytical Instruments, Madison, WI) was used for the collection of all spectra. A 250 W tungsten-halogen lamp, a calcium fluoride beamspilitter and a cryogenically cooled indium antimonide (InSb) detector were used. A multilayer optical interference filter (Barr Associates, Westford, M A ) was used to confine the spectral range of 5000 - 4000 cm" . The sample cell was a rectangular infrasil quartz cell with a path length of 1 mm (Wilmad Glass Co., Buena, NJ). Temperature was controlled by placing the sample cell in an aluminumjacketed cell holder in conjunction with a V W R 1140 refrigerated temperature bath (VWR Scientific, Chicago, IL). Temperatures were measured with a copperconstantan thermocouple probe placed directly in the sample solution and reading were obtained from an Omega Model 670 digital meter (Omega Inc., Stamford, CT). 1

118

BIOSENSOR AND C H E M I C A L SENSOR T E C H N O L O G Y

Asparagine and glutamine were purchased from Sigma Chemical Co. (St. Louis, MO). Sodium bicarbonate and sodium phosphate monohydrate were obtained from Aldrich Chemical Company, Inc. (Milwaukee, WI). All aqueous solutions were prepared in deionized water purified with a Milli-Q Reagent Water System (Millipore, Bedford, MA). Sixty-six different binary mixtures were prepared by randomly mixing different milligram quantities of asparagine and glutamine in 10 ml of buffers. The buffer was composed of 4.17 m M bicarbonate and 8.44 mM phosphate, and the p H was set at 6.35. A l l measurements for a particular solution were completed in less than one hour after the solution was prepared. Fresh samples were used to avoid complications caused by hydrolysis of these amino acids. Spectra Collection and Processing. Spectra collection started after placing the sample in the sample holder and waiting for a two minute equilibration period. Spectra were collected as double-sided interferograms of256 co-added scans. After triangularly apodized, the interferograms were Fourier transformed to generate single beam spectra with a spectral resolution of 1.9 cm" . Single beam spectra were later transferred to an Iris Indigo computer (Silicon Graphics, Inc., Mountain View, CA) for processing. Three single spectra were collected sequentially for each sample solution without moving the sample. Blank buffer solutions were used for the collection of background single beam spectra. After every fourth sample measurement, a new background spectrum was collected and used as the reference spectrum for subsequent sample spectra. Each single beam sample spectrum was ratioed against the corresponding single beam reference spectrum Absorbance spectra were generated as the negative logarithm of the ratioed spectra. A total of 197 absorbance spectra were obtained with the 66 sample solutions. A l l the computer software used for spectral processing was obtained from Professor Gary Small in the Center for Intelligent Instrumentation in the Department of Chemistry at Ohio University (Athens, OH). Subroutines for Fourier filtering and PLS regression were obtained from the IMSL software package (IMSL, Inc., Houston, TX). 1

Results and Discussion Absorption Bands. Success of remote NIR spectroscopic sensing depends on the identification and isolation of unique spectral bands for the targeted analytes. These analyte bands must be distinguishable from those of water and other analytes in the sample. Water has large absorption bands with peak maxima at 6876, 5267, and 3800 cm" in the NIR (12). The spectral region of 5000-4000 cm" corresponds to an optical window between two large water absorption bands with a minimum around 4500 c m . After taking into account the buffer background, the spectral characteristics of asparagine and glutamine can be compared. Asparagine and glutamine, like other amino acids, have discernible absorption bands around 4390, 4570, and 4700 cm" . The first band corresponds to the combination bands of aliphatic C-H. The second and third bands correspond to the combination bands of amine N - H . The exact 1

1

1

1

12. ZHOU ET AL.

Selective Measurement of Glutamine and Asparagine

119

position and width of each band varies for the different amino acids depending on the chemical composition and structure of the side chains. The chemical differences between glutamine and asparagine result in rather subtle differences in their NIR spectra (see Figure 1). The largest difference is the position and width of the low frequency absorbance bands. This band is centered at 4393 cm" for glutamine and 4369 cm" for asparagine. In addition, the asparagine band is wider than that of glutamine. These differences are related to the additional methylene group of glutamine which corroborates earlier findings that NIR bands around 4400 cm" are associated with the first combination bands of C - H vibrational transitions (8, 11, 15). All the other spectral features are similar in appearance with only minor differences in position, magnitude and width. 1

1

1

P L S Regression. The whole data set representing 197 spectra from 66 samples was divided into two data sets with one for calibration and one for prediction. The calibration data set was composed of all spectra corresponding to 50 samples which were selected randomly from the whole data set. A total of 149 spectra were in the calibration set. The remaining spectra (48 spectra from 16 samples) were placed in the prediction set. As is reported (10), when applying the PLS algorithm to correlate spectral variations with concentration variations, there should not exist any correlation between the concentrations of the two analytes. The combinations used in this study were built by random selection. Figure 2 presents the correlation plot between the concentrations of asparagine and glutamine in the sample solutions. This plot clearly shows there is no correlation between these two compounds. The r-square value of a linear regression analysis of the entire data set is 0.001. The corresponding r-square values are 0.034 and 0.266 for the calibration and prediction data sets, respectively. For PLS regression, the spectral range and number of PLS factors are the two crucial parameters. Five spectral ranges were tested for both asparagine and glutamine. They are 4800-4250, 4700-4320, 4650-4320, 4450-4320, and 47004450 cm" . The first three ranges incorporate both the 4570 and 4390 cm" absorption bands, but differ in the width of the spectral range. The fourth and fifth ranges focus on the 4390 and 4570 c m absorption bands, respectively. The optimum number of PLS factors were determined for each spectral range and each analyte. Table I presents the results of this calculation. For each spectral range and each analyte, the number of factors listed corresponds to the minimum mean standard error of prediction (SEP) (13, 14). A n example of our finding is presented in Figure 3a which shows the standard errors of calibration (SEC) and prediction (SEP) as functions of the number of PLS factors used for glutamine with the spectral range of 4800-4250 cm" . Initially, both the SEC and SEP drop sharply as more of the spectral variation in the calibration data set is incorporated into the model. Little improvement in either the SEC or SEP is observed after 8 factors. As expected, the SEC continues to drop and the SEP begins to increase as the number of factors is increased further. The slight increase in the SEP indicates that the data is overmodelled with too many factors. The effect of spectral range is typified by the curves presented in Figure 3b which 1

1

1

1

120

BIOSENSOR AND CHEMICAL SENSOR T E C H N O L O G Y

0.0040 0.0035 Asparagine 8 α

0.0030

I

0.0025

•B 0.0020 Η 0.0015

o.ooio Η 0.0005

JL 4800

_L 4700

JL 4600

4500

4400

4300

1

Wavenumber (cm" ) Figure 1. NIR spectra of 10 m M asparagine and 10 mM glutamine in phosphate buffer.